Due to the rapid process of urbanization, there is an increasing demand for detecting building changes over time using very high-resolution (VHR) images. Traditional two-dimensional (2-D) change detection methods are limited due to the image perspective variation and illumination discrepancies. One current trend for building detection combines the use of orthophotos and digital surface models (DSMs), because of its robustness against false changes, as well as its capability of providing volumetric information. In this paper, we propose an object-based three-dimensional (3-D) building change detection framework based on supervised classification, which makes use of the height, spectral, and shape information in a combined fashion with object-based analysis. The proposed method follows the following steps: First, a synergic mean-shift segmentation method is applied on the orthophoto with the constraints of the DSM, which derives segments with homogenous spectrum and height.
In a second step, the segments are classified with a hybrid decision tree and SVM approach, and then the segments of the building class are merged as building objects for change detection. An initial change indicator (CI) is then computed for each building object concerning height and spectral information. Finally, an adaptive CI updating strategy based on segment overlapping is proposed and the traffic light system based on a dual threshold is used to identify the change status of each building as “change,” “no-change,” and “uncertain change”. The experimental results on scanned aerial stereo images have demonstrated that our proposed framework is able to achieve high-detection accuracy on images with limited spectral quality.